Efficient Online Controller Tuning for Omnidirectional Mobile Robots Using a Multivariate-Multitarget Polynomial Prediction Model and Evolutionary Optimization

摘要

The growing reliance on mobile robots has resulted in applications where users have limited or no control over operating conditions. These applications require advanced controllers to ensure the system's performance by dynamically changing its parameters. Nowadays, online bioinspired controller tuning approaches are among the most successful and innovative tools for dealing with uncertainties and disturbances. Nevertheless, these bioinspired approaches present a main limitation in real-world applications due to the extensive computational resources required in their exhaustive search when evaluating the controller tuning of complex dynamics. This paper develops an online bioinspired controller tuning approach leveraging a surrogate modeling strategy for an omnidirectional mobile robot controller. The polynomial response surface method is incorporated as an identification stage to model the system and predict its behavior in the tuning stage of the indirect adaptive approach. The comparative analysis concerns state-of-the-art controller tuning approaches, such as online, offline robust, and offline non-robust approaches, based on bioinspired optimization. The results show that the proposal reduces its computational load by up to 62.85% while maintaining the controller performance regarding the online approach under adverse uncertainties and disturbances. The proposal also increases the controller performance by up to 93% compared to offline tuning approaches. Then, the proposal retains its competitiveness on mobile robot systems under adverse conditions, while other controller tuning approaches drop it. Furthermore, a posterior comparison against another surrogate tuning approach based on Gaussian process regression corroborates the proposal as the best online controller tuning approach by reducing the competitor's computational load by up to 91.37% while increasing its performance by 63%. Hence, the proposed controller tuning approach decreases the execution time to be applied in the evolution of the control system without deteriorating the closed-loop performance. To the best of the authors' knowledge, this is the first time that such a controller tuning strategy has been tested on an omnidirectional mobile robot.